CHANGE DETECTABILITY IN BAYESIAN SETTING

This paper presents an approach to assess minimum detectable parameter changes based on Bayesian inference and the concept of highest posterior density interval. The method is developed for structural health monitoring problems, where observations of system outputs are used to infer knowledge about random system inputs. The analysis is based on linear Bayesian filters and the parameter change is defined as the shift in the mode of the distribution. For proof of concept, the framework is applied to a case study, that is, a numerical model of an offshore tower affected by marine growth. A functional Kalman filters is used to predict the minimum detectable changes, and for cross-validation, the prediction is validated based on a Markov Chain Monte Carlo simulation. The results show that, as long as the parameter can be satisfactorily identified, the minimum detectable parameter changes can also be accurately predicted. One of the advantages of the approach is that the minimum detectable parameter change can be predicted based on observations on the unchanged system. Moreover, it can be applied to a wide range of observed features, damage scenarios, and linear or non-linear systems. In contrast to existing approaches, the presented version is not restricted to Gaussian distributions.